Göttingen 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 76: Data, AI, Computing, Electronics VII (Generative AI, MC Generators)

T 76.4: Vortrag

Donnerstag, 3. April 2025, 17:00–17:15, VG 2.101

PointL2LFlows: How to generate Hadronic showers in ECal and HCal with CNFs — •Thorsten Buss — Institut für Experimentalphysik, Universität Hamburg, Germany

In collider experiments, Monte Carlo (MC) simulations are the essential tool for comparing experimental findings with theory predictions. However, they have a high computational demand, and future developments, such as higher event rates, are expected to increase this demand beyond availability.

Generative models provide a way of augmenting MC simulations, speeding them up, and overcoming this bottleneck. Recent works have successfully applied this approach to electromagnetic showers in electromagnetic calorimeters (ECal) and to pion showers in low-granular homogeneous calorimeters. However, applying it to pion showers developing in a highly granular ECal and continuing in a highly granular HCal remains a challenge due to their rich substructure.

This work shows how point-cloud-based continuous normalizing flows (CNF) can jointly generate pion showers in ECal and HCal. As in our L2LFlows model for EM showers, we generate one calorimeter layer at a time conditioned on the previous layers. This reduces the size of the point clouds reducing computational costs and making it easier for the model to focus on the most important structures in the showers.

Keywords: Continuous Normalizing Flows; Detector Simulation; Hadron Shower; Calorimeter